According to a Financial Times report on Feb 3, 2023, Google invests about $300mn in Anthropic, a generative-AI startup, to take about 10% of its ownership. Anthropic, emphasizing AI safety and research, aims at building reliable and interpretable AI systems. In developing Claude (not yet released), their major product rivaling ChatGPT, they used a list of principles serving as the “constitution” (the technique is named constitutional AI). The principles are about maximizing positive impact, enabling freedom of choice and preventing the system from making harmful advice. Anthropic has been using Google’s cloud computing services.
Google’s investment echoes Microsoft’s recent $10bn investment in OpenAI, the company behind ChatGPT. Microsoft is expected to integrate OpenAI’s abilities into its services, such as the Bing search engine. Competing with Microsoft, Alphabet is strengthening its push in AI with the move with Anthropic, the collaboration with other startups such as Cohere, the development of LaMDA-based chatbot, and a shift towards language modelling in DeepMind. Other companies, such as Meta and Amazon, are actively developing their own AI capabilities as well. In China, Baidu.com is expected to add AI to search. For now, Alphabet and Microsoft seem equipped to become two powerhouses in large language model (LLM) related technologies. Their chatbots, based on LaMDA and GPT-3.5 respectively, are state-of-the-art, neck and neck in performance. However, not everyone is convinced that large language model is the future. Notably, Yann LeCun, the chief AI scientist at Meta, does not think highly of LLM. He believes some major breakthroughs are needed before AI can approach human-level capabilities, such as developing baby-like ability to learn how the world works and the ability to predict how actions affect the world. If Meta develops an AI system totally different from large language models, we should not be surprised.
More fierce competition in AI is on the way.
Further Possibilities
1. Use search data to improve generative AI
2. Connecting generative AI and Google Scholar to facilitate academic research
3. Two AI and startup ecosystems, supported by Alphabet and Microsoft respectively, emerge
4. A new foundation model, different from large language models, appears from other companies
5. Adding generative AI capabilities into Google Docs
